Item

Abstract

In recent years many genome-wide association (GWA) studies have been performed. Many of these have been successful in identifying loci associated with complex diseases. Until now, these results failed to fully explain the heritability of many of the diseases. Searching for interactions between Single Nucleotide Polymorphisms (SNPs) is one of the many possible explanations to the missing heritabili- ty. However, the computational time needed for testing all pairs of SNPs is proportional to the square of the number of SNPs, translating into months of CPU time. We therefore evaluated GLIDE [1] which makes use of the com- putational power of consumer-grade graphics cards (GPUs) to detect inter- actions via linear regression. We present our first experiments with GLIDE for which we analysed the HDL levels in 3,000 individuals of the Rotterdam Study. The first results show that GPUs can indeed be used for fast genome-wide analysis of SNP-SNP interactions. We found multiple regions showing genome-wide significant interactions which are currently being investigated. However, occasionally problems occur using imputed data hampering meta-analysis and replicati- on. Further developments are currently ongoing to overcome this issue and will also be presented.